jeremygracey-ai/provenance-corpus
收藏资源简介:
该数据集是Provenance Corpus — OpenStax Anatomy & Physiology 2e (page images),包含OpenStax《解剖学与生理学》第2版教科书的页面图像,共1,347页,每页以PNG格式存储(分辨率为150 DPI)。它用作Provenance演示系统的检索语料库,Provenance是一个基于引用的检索增强生成(RAG)系统,能够验证每个声明对应的确切页面图像。数据集通过多模态模型(Cohere Embed v4)进行嵌入,支持文本查询和页面图像在共享向量空间中的检索,并通过余弦相似度匹配。检索到的图像用于Claude模型的回答和逐项声明验证。页面通过ID(如anatomy-physiology-2e#p{N})引用,并可通过特定URL访问。数据来源于OpenStax的CC BY 4.0许可内容,数据集同样遵循该许可。
This dataset is the Provenance Corpus — OpenStax Anatomy & Physiology 2e (page images), containing page-image renders of the OpenStax Anatomy & Physiology 2e textbook (1,347 pages), with each page stored as a PNG at 150 DPI. It serves as the retrieval corpus for the Provenance demo—a citation-grounded RAG system that verifies every claim against the exact page image that proves it. The images are embedded using Cohere Embed v4 (a multimodal model where query text and page images share one vector space) and retrieved by cosine similarity. Retrieved page images are then passed to Claude for answering and per-claim verification. Pages are referenced by IDs (e.g., anatomy-physiology-2e#p{N}) and accessible via specific URLs. The source is OpenStax content under CC BY 4.0 license, and the dataset is redistributed under the same license.




